Representation learning approaches provide remarkable results but are often based on complex architectures requiring huge datasets and unsustainable computations. Despite this gargantuan amount of data and compute, state-of-the-art methods often generalize poorly in practice due to slight changes in between the training and test distribution. Causal machine learning offers an elegant approach to study this problem but it does not scale as well as modern deep learning. Our goal in this project is to explore how causal structure in the data distribution can inform the design of novel forms of inductive bias, making AI models easier to train, more sustainable, and more reliable in solving real world problems.

Funded by PON-REACT EU.

Collaboration with

UniGe | LCSL @ MaLGa
Francesco Locatello, Amazon AWS

Shearlets are a multi-resolution analysis framework with many suitable properties for it to be applied to the analysis of images and videos. Among these, we mention its ability in characterizing anisotropic structures, and in enhancing signal singularities. For these reasons, we have adopted it in the detection and descriptions of keypoints in images and image sequences. Its robustness to noise (including motion blur and compression artifacts), makes it suitable to address different applications in the signal processing domain.

References

D Malafronte, E De Vito, F Odone, Space-time signal analysis and the 3D Shearlet Transform, Journal of Mathematical Imaging and Vision 2018

MA Duval-Poo, N Noceti, F Odone, E De Vito, Scale invariant and noise robust interest points with shearlets, IEEE Transactions on Image Processing 26 (6), 2017

MA Duval-Poo, F Odone, E De Vito, Edges and corners with shearlets, IEEE Transactions on Image Processing, 2015

Collaboration with

UniGe | CHarML @ MaLGa

Complex network topologies are the fundamental substrate supporting complex brain functions. In vitro neural networks are a validated experimental model for studying the mechanisms governing the formation, and organization of neuronal cell assemblies. The project aims in designing machine learning algorithms to analyze electrophysiological signals acquired from in vitro neural networks by means of micro electrodes array.

References

Poli, Daniele, Vito P. Pastore, and Paolo Massobrio. Functional connectivity in in vitro neuronal assemblies. Frontiers in neural circuits 9 (2015): 57.

Collaboration with

UNIGE NBT Lab (Sergio Martinoia, Paolo Massobrio, Michela Chiappalone)

The relationship between the genotype, the genetic instructions encoded into a genome, and phenotype, the macroscopic realization of such instructions, remains mostly uncharted. This project aims to develop machine learning tools and frameworks of analysis in the attempt of relating changes of cellular morphology to genetic modification. We are exploiting unsupervised learning and transfer learning tools to deal with the task challenges, focusing on the yeast vacuoles as test organelle.

References

Vito Paolo Pastore, Ashwini Oke, Sara Capponi, Daniel Elnatan, Jennifer Fung, Simone Bianco, Phenotype to genotype mapping using supervised and unsupervised learning. bioRxiv 2022.03.17.484826

Collaboration with

UCSF Department of Obstetrics, Gynecology and Reproductive Sciences (Jennifer Fung, Ashwini Oke)
IBM Research Almaden, Cellular engineering group (Sara Capponi)
Altos Labs, Simone Bianco